mlsanityData to 14 Oct 2026

Scaling Laws for Forgetting When Fine-Tuning Large Language Models

arXiv 2401.05605 ↗First posted 11 Jan 2024cs.CL cs.LGcited by 38

Not on the map. Why?

Abstract (arXiv)

We study and quantify the problem of forgetting when fine-tuning pre-trained large language models (LLMs) on a downstream task. We find that parameter-efficient fine-tuning (PEFT) strategies, such as Low-Rank Adapters (LoRA), still suffer from catastrophic forgetting. In particular, we identify a strong inverse linear relationship between the fine-tuning performance and the amount of forgetting when fine-tuning LLMs with LoRA. We further obtain precise scaling laws that show forgetting increases as a shifted power law in the number of parameters fine-tuned and the number of update steps. We also examine the impact of forgetting on knowledge, reasoning, and the safety guardrails trained into Llama 2 7B chat. Our study suggests that forgetting cannot be avoided through early stopping or by varying the number of parameters fine-tuned. We believe this opens up an important safety-critical direction for future research to evaluate and develop fine-tuning schemes which mitigate forgetting

Key facts. Cited by 38 papers in the map (data through 14 Oct 2026). Builds on 33 of its 45 references that are in the map. First posted on arXiv 11 Jan 2024.

Builds on

33 of 45 references.

References · 33

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2022Training language models to follow instructions with human feedbackcited by 15,971
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  4. 2020Scaling Laws for Neural Language Modelscited by 6,586
  5. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,886
  6. 2016Overcoming catastrophic forgetting in neural networkscited by 4,787
  7. 2021On the Opportunities and Risks of Foundation Modelscited by 4,033
  8. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,780
  9. 2016Pointer Sentinel Mixture Modelscited by 3,096
  10. 2022Training Compute-Optimal Large Language Modelscited by 3,019
  11. 2016iCaRL: Incremental Classifier and Representation Learningcited by 2,333
  12. 2017Gradient Episodic Memory for Continual Learningcited by 1,898
21 more
  1. 2016Progressive Neural Networkscited by 1,361
  2. 2018Continual Lifelong Learning with Neural Networks: A Reviewcited by 1,330
  3. 2021BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-modelscited by 1,245
  4. 2018Experience Replay for Continual Learningcited by 1,095
  5. 2018Adafactor: Adaptive Learning Rates with Sublinear Memory Costcited by 905
  6. 2022Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learningcited by 896
  7. 2013An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networkscited by 722
  8. 2023An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuningcited by 639
  9. 2020Scaling Laws for Autoregressive Generative Modelingcited by 522
  10. 2016Expert Gate: Lifelong Learning with a Network of Expertscited by 372
  11. 2017Measuring Catastrophic Forgetting in Neural Networkscited by 360
  12. 2022Unified Scaling Laws for Routed Language Modelscited by 253
  13. 2023A Rank Stabilization Scaling Factor for Fine-Tuning with LoRAcited by 243
  14. 2018Selective Experience Replay for Lifelong Learningcited by 241
  15. 2018Reinforced Continual Learningcited by 186
  16. 2022Delta Tuning: A Comprehensive Study of Parameter Efficient Methods for Pre-trained Language Modelscited by 180
  17. 2019Uncertainty-based Continual Learning with Adaptive Regularizationcited by 146
  18. 2019Mixout: Effective Regularization to Finetune Large-scale Pretrained Language Modelscited by 141
  19. 2022The Challenges of Continuous Self-Supervised Learningcited by 36
  20. 2017Neural Domain Adaptation for Biomedical Question Answeringcited by 30
  21. 2017Representation Stability as a Regularizer for Improved Text Analytics Transfer Learningcited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 7; 2025: 12; 2026: 19.

38 papers.

Sort:

By area:

  1. 2024LoRA Learns Less and Forgets Lesscited by 328
  2. 2024MiLoRA: Harnessing Minor Singular Components for Parameter-Efficient LLM Finetuningcited by 94
  3. 2024Large Scale Transfer Learning for Tabular Data via Language Modelingcited by 76
  4. 2025How Do Large Language Monkeys Get Their Power (Laws)?cited by 43
  5. 2025Scaling Laws for Forgetting during Finetuning with Pretraining Data Injectioncited by 35
  6. 2025Efficient Prediction of Pass@k Scaling in Large Language Modelscited by 16
  7. 2024Human-inspired Perspectives: A Survey on AI Long-term Memorycited by 12
  8. 2024TeamLoRA: Boosting Low-Rank Adaptation with Expert Collaboration and Competitioncited by 10
  9. 2025Aegis: Taxonomy and Optimizations for Overcoming Agent-Environment Failures in LLM Agentscited by 10
  10. 2024Demystifying Language Model Forgetting with Low-rank Example Associationscited by 8
  11. 2026Can Muon Fine-tune Adam-Pretrained Models?cited by 7
  12. 2026Learning, Fast and Slow: Towards LLMs That Adapt Continuallycited by 6
26 more
  1. 2025The Devil in the Details: Emergent Misalignment, Format and Coherence in Open-Weights LLMscited by 5
  2. 2024Fine-tuning Large Language Models with Limited Data: A Survey and Practical Guidecited by 4
  3. 2025A Comparative Analysis of LLM Adaptation: SFT, LoRA, and ICL in Data-Scarce Scenarioscited by 4
  4. 2025Improving Rule-based Reasoning in LLMs using Neurosymbolic Representationscited by 3
  5. 2026MixSD: Mixed Contextual Self-Distillation for Knowledge Injectioncited by 3
  6. 2026MemSFT: Mitigating Alignment Tax with an External Parametric Memorycited by 3
  7. 2025Evaluating the Robustness of Chinchilla Compute-Optimal Scalingcited by 2
  8. 2026Least but not Last: Fine-tuning Intermediate Principal Components for Better Performance-Forgetting Trade-Offscited by 2
  9. 2026Fine-tuning MLLMs Without Forgetting Is Easier Than You Thinkcited by 2
  10. 2026The Intruder Threshold: A Spectral Law for LoRA Fine-Tuningcited by 2
  11. 2026Fine-Tuning Without Forgetting via Loss-Adaptive Learning Ratescited by 1
  12. 2026The Future of Facts: Tracing the Factual Generation-Verification Gapcited by 1
  13. 2025AILoRA: Function-Aware Asymmetric Initialization for Low-Rank Adaptation of Large Language Modelscited by 0
  14. 2025R2R: A Route-to-Rerank Post-Training Framework for Multi-Domain Decoder-Only Rerankerscited by 0
  15. 2025SignRoundV2: Toward Closing the Performance Gap in Extremely Low-Bit Post-Training Quantization for LLMscited by 0
  16. 2025CureAgent: A Training-Free Executor-Analyst Framework for Clinical Reasoningcited by 0
  17. 2026PLATE: Plasticity-Tunable Efficient Adapters for Geometry-Aware Continual Learningcited by 0
  18. 2026Awakening the Sleeping Agent: Lean-Specific Agentic Data Reactivates General Tool Use in Goedel Provercited by 0
  19. 2026Teaching LLMs Brazilian Healthcare: Injecting Knowledge from Official Clinical Guidelinescited by 0
  20. 2026Smarter by the Moment: Environment-Driven Dynamic Policies for Continual LLM Improvementcited by 0
  21. 2026SafeTune: A Unified Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMscited by 0
  22. 2026ViSTA: A Simple Bridge Extends Visual Alignment to Clinical Time-Series Understanding in Multimodal LLMscited by 0
  23. 2026When Can First-Order Models of Fine-Tuning Bound Forgetting?cited by 0
  24. 2026Learning from the Gap Between Pass@K and Pass@1cited by 0
  25. 2026MetaSteer: Context-Conditioned, nonlinear Steering via Attention-Projection Adaptationcited by 0
  26. 2026Continually Post-training Small Models via Synthetic Self-distillationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Leaner Training, Lower Leakage: Revisiting Memorization in LLM Fine-Tuning with LoRA
  2. 2026Mechanistic Analysis of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning
  3. 2024Examining Forgetting in Continual Pre-training of Aligned Large Language Models
  4. 2025Catastrophic Forgetting in LLMs: A Comparative Analysis Across Language Tasks
  5. 2026Talking to Yourself: Defying Forgetting in Large Language Models
  6. 2026CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training
  7. 2024Self-Distillation Bridges Distribution Gap in Language Model Fine-Tuning
  8. 2026Are Parameter-Efficient Fine-tuning Methods Really Different?